Re: log file sync vs log file parallel write
Date: Fri, 3 Oct 2008 05:13:40 -0700 (PDT)
First of all I am not a newbie.
I know what the meaning of the events are.
I googled and read books that are available on performance tuning.
The only thing I did not understand is that the average write to the
redo logs is about 10ms.
But why should the log file sync be 10 times as high.
In the mean time I continued reading and came across a not of Jonathan
Lewis about bug 2669566
So I think I may be hitting this bug.
You should not judge that fast.
On 3 okt, 13:22, sybrandb <sybra..._at_gmail.com> wrote:
> On 3 okt, 12:31, HansP <hans-peter.sl..._at_atosorigin.com> wrote:
> > Hello all,
> > While investigating performance problems in a database I discovered
> > that the 'log file sync' wait event is almost always in the top 5.
> > (I am looking at the statspack datat)
> > Calculating the ratio of the average wait time of the 'log file sync'
> > and the 'log file parallel write' gives a ratio of 10.
> > Does someone know why this may happen?
> > Regards Hans-Peter
> First of all,
> 'Log file sync' is associated with commit.
> If you see this event in the top 5, you are committing way too often
> (maybe for every individual record) and it is
> developer beating time!!! They don't know what a transaction is, and
> they need to be whipped for that.
> Secondly 'log file parallel write' obviously deals with flushing out
> the redo log buffer to disk. This means your disks are too slow, or
> you have an I/O bottleneck.
> Please note there is plethora of resources on 'log file sync' online,
> particularly in this very forum.
> Please force yourself being less lazy and use Google, your friend, to
> search this and other forums, instead of asking this boring FAQ AGAIN.
> The basic problem with many newbie DBAs is not they are short of
> knowledge, the basic problem is they are too lazy, and want to be
> spoon fed.
> That won't bring you anywhere.
> Sybrand Bakker
> Senior Oracle DBA
Received on Fri Oct 03 2008 - 07:13:40 CDT